YouTube Video Script: "Who AI Search Actually Cites" (EPISODE 1)

For: SEO Believer (channel-building, ongoing research series) Series: AI Search, Studied. Episode 1 of an ongoing series where our team runs real experiments on what the AIs actually recommend, then reports what happened. Delivered by: Annette Thompson's AI avatar Source study: research/studies/who-ai-search-actually-cites-PUBLIC.md (anonymized public version) Strategy basis: research/studies/youtube-channel-strategy.md (2026 best practices, with sources) Format: Two-column script (visual / b-roll + voiceover) with timestamps + YouTube chapters Runtime target: 8 to 10 minutes (target ~9). This is the data-backed sweet spot for educational/authority content. We are not padding; the added time goes to a sharper hook, section micro-hooks, a mid-roll re-hook, an explicit subscribe moment, and an Episode 2 tease, all of which lift retention rather than dilute it. Voiceover word count: ~1,400 words (see count at end), reading to roughly 8.5 to 9.5 minutes at a natural avatar pace.

Voice notes for the avatar read: warm, plain-spoken, unhurried. Use contractions. No em-dashes anywhere. Say "AI," not "engine," except for the proper term "Generative Engine Optimization (GEO)" and "search engine" only when we mean Google or Bing. We speak as a team ("we," "our team"), not a solo operator. Do not stress or lean on the word "believers." Reference Colorado mountain visuals only with a clear blue sky and white clouds, never a white or overcast sky.

Retention design (why this script is shaped the way it is): a video that holds 70% of viewers past 0:30 gets recommended to far more new viewers than one that loses 40% in the same window. So the first 8 seconds make a concrete, stakes-first claim; the next 17 seconds state exactly what the viewer will know by the end and name the series; every finding opens with a one-line micro-hook so the retention curve re-lifts at each section boundary; and a mid-roll open loop holds people toward the most surprising finding. New channels win subscribers from search first, so the title, the spoken opening, and the description all front-load the exact question a business owner types.

---

THE SCRIPT (two columns)

TIME	VISUAL / B-ROLL	VOICEOVER
0:00 - 0:08	COLD OPEN. No logo, no intro. Tight screen-recording: a cursor types into ChatGPT "best med spa in Boulder Colorado" and hits enter. The answer animates in, and we highlight a real business name appearing in the response. One beat of silence on the name.	Right now, an AI is recommending a business in your town, by name, to someone who's about to spend money. The only question that matters is whether that name is yours.
0:08 - 0:25	Quick montage, three screen-recordings side by side: ChatGPT, Perplexity, and Gemini each answering a "best [category] near me" question, each naming different businesses. Names blurred lightly for courtesy. Brand bug fades in bottom corner: SEO Believer. Small series badge animates in: "AI Search, Studied / Ep. 1."	Here's what we're going to show you in the next few minutes. We asked six different AIs the exact questions your future customers are already asking them, and we wrote down who got named and who got left out. By the end you'll know whether your industry is still open, which single page can win you the most AI mentions for the least money, and who your real competitor actually is. This is episode one of a series where our team keeps running these experiments and reporting what we find.
0:25 - 0:42	Drone footage we own: the Boulder Flatirons at golden hour under a clear blue sky with a few white clouds, slow push-in. Then a clean title card on cool white: "Who AI Search Actually Cites. A study of 30+ high-intent local queries. Front Range, June 2026." Antique gold underline draws in.	This is real data, captured live across six different AIs, in eight kinds of local business, here on the Colorado Front Range. It's a small study, and we're honest about that. But it already overturned a few things we believed going in. Quick favor before we start. If you run a local business, subscribe, because we're going to keep doing this every few weeks for a different set of industries. Now let me walk you through what we found.
0:42 - 1:15	CHAPTER: Finding 1. Animated graphic: a single row of six AI logos (ChatGPT, Perplexity, Claude, Grok, Gemini, Google AI Overviews) light up one at a time. Below each, a different business name card flips up, all different. Then a big animated question mark settles over the row.	We started with the question every owner asks us. "Do I really need to worry about AI search yet?" Here's the first finding, and it's the most useful thing we learned. There is no stable winner. Ask six AIs the same "best med spa in Boulder" question, and you get a different lead name almost every time. One AI named a business that another one left out completely. Adding Google's Gemini made it more fragmented, not less. It named a set that barely overlapped anyone else.
1:15 - 1:38	The six business-name cards rearrange into an "AI agreement matrix" grid (rows = businesses, columns = AIs, filled cells scattered, no full row). Camera pulls back to show how patchy it is. A glowing empty cell pulses.	And here's why that should make you sit up, not tune out. That instability is the entire opportunity. Think about it. If one business already owned every AI's answer, there'd be nothing left to win. Because the named set is scattered across all these AIs, a business can move from being named by one of them to being named by all of them. That movement is something our team can actually engineer.
1:38 - 2:15	CHAPTER: Finding 2. Screen-recording: type "dental implants Boulder cost" into Perplexity. Answer loads. Highlight a price figure. Cut to Claude, same query, same price highlighted. Cut to Grok, same. Then a "$3,150" number-callout animates large on screen with a gold ring.	The second finding is the one we keep coming back to, because it's where the cheapest win hides. On pricing questions, one source often feeds every AI. We asked "dental implants Boulder cost." Perplexity, Claude, and Grok all anchored the price to the same source. Not a big directory. Not a comparison site. One local practice's well-built pricing page, with one clean, specific, quotable number. Every AI reached for it because it was the easiest number to quote.
2:15 - 2:42	"One source feeds the answer" diagram animates: a single pricing-page node in the center, arrows shooting out to four AI logos. Off to the side, a Gemini logo connects with a dashed line and a small label "generic ranges only." A second empty node labeled "the open seat" pulses next to it.	Then Google's own AI Overview confirmed it from the inside, naming that same practice's number. So four AIs plus Google's overview all converge on one page. Gemini was the honest exception. It stayed generic and named no local source at all, and that tells us exactly where the next work goes. But the lesson is blunt. There's room for exactly one more source. And being that source means getting quoted by most of the AIs at once.
2:42 - 3:33	CHAPTER: Finding 3. The vertical-openness ladder graphic builds bar by bar, top to bottom, green at the top fading to deep forest then to a muted red at the bottom. Each bar labels a vertical as it slides in: dental implant cost, braces cost, med spa, cosmetic dentistry, plastic surgery, LASIK, fertility, personal injury law.	The third finding answers the question I know you're already asking. "Okay, but can I win?" The honest answer is, it depends entirely on your line of work. We checked each kind of business and asked one thing. When someone asks an AI to recommend a business like yours, does it already have a fixed set of favorites, or is there still room for a newcomer? Look at this ladder. At the top, in green, dentists and orthodontists asking about cost. The easiest wins we found, and cheap. Med spas and cosmetic dentists, still open. Then it tightens. LASIK is mostly closed. And at the bottom, in red, fertility clinics and personal injury law. Already locked. So the real question isn't "can anyone win at AI search?" It's "can I win, in my industry, in my town?"
3:33 - 3:48	MID-ROLL RE-HOOK. Quick, confident cut back to the avatar for one beat, or a clean interstitial card on cool white: "Still coming: the one we got wrong." A small "3 of 6 findings" progress marker ticks forward.	Stay with me, because the next finding is the one we got most wrong going in, and it changes who you think you're competing against.
3:48 - 4:28	CHAPTER: Finding 4. Map graphic of the US with six city pins lighting up: Boulder, Denver, Phoenix, Nashville, Tampa, Austin. A "telehealth brand" logo placeholder tries to appear on each pin and gets a soft gray X. Then local clinic pins glow green. Nashville and Boulder pulse "open"; Austin and Tampa show "locked."	The fourth finding surprised us most. We expected AI to push the big national telehealth brands for hormone-therapy searches. It didn't. Not once. Across all six metros and every AI we tested, the answers were local, in-person clinics. The real competitor wasn't a national app at all. It was the multi-location franchise planting flags in your local map. Where a franchise had several locations, it crowded the answer. Where it had one or none, like here in Boulder, local independents owned the ground. That's defensible, if you move before the chain does.
4:28 - 5:08	CHAPTER: Finding 5. Signal-by-AI graphic: four columns labeled "Reviews and listings," "Clear, well-labeled pages," "Reddit and forum presence," "Well-known name in Google's index." AI logos drop under the signal each rewarded. Lines connect to show one business needing to satisfy all four.	The fifth finding is the one that saves you from getting sold a bad fix. Different AIs reward different things. Grok leaned on Reddit and forum chatter, surfacing people that real humans recommended in threads. Perplexity and ChatGPT leaned on review counts and listings. Claude rewarded a clean, well-organized page. Google's Overviews rewarded dated, number-dense pages. And Gemini rewarded being a well-known, established name in Google's index. That's five different things to satisfy, with one business. Which is why anyone selling you a single one-trick fix is really just selling you one AI's answer.
5:08 - 5:43	CHAPTER: Finding 6. Split-screen "AI Overview trigger rate" graphic. Of 20 Google queries, an animated counter fills: 11 rendered an overview, 9 returned the local map 3-pack. The 11 group under "cost questions." The 9 group under "fragmented best-of questions." Annotation types in.	And the sixth finding is the clearest pattern in the Google data. We ran twenty queries through Google. An AI Overview showed up on about fifty-five percent of them, and the split wasn't random. Every cost question triggered an overview. But the fragmented "best of" questions didn't. Google handed back the old map pack instead. The read is simple. When there's a settled number or a settled set of names, Google summarizes it. When the field is genuinely up for grabs, it declines, and gives you the map. That tells you which game you're even playing.
5:43 - 6:35	CHAPTER: What this means for your revenue. Clean white screen, antique-gold number "1." Soft b-roll: an empty waiting room, a phone that doesn't ring, a calendar with open slots. Then "2." with the $3,150 page reappearing small. Sage-green checkmark animates on.	So let's make this practical. What does it all mean for your revenue? First, when AI leaves you out, nothing breaks visibly. No alert fires. You just quietly stop being considered by the growing share of buyers who ask an AI before they ask a friend. The customers you lose this way never knew you existed, so you never see them. The damage you can't see is the dangerous kind. Second, the cheapest win is the pricing page. If you're in a category where one source feeds the cost answer, a single well-built page can make you the next one the AIs quote, across most of them at once. That's the highest-probability, lowest-cost win in the entire study.
6:35 - 7:12	Number "3." with the vertical ladder flashing back briefly, then "4." with a franchise logo placeholder being out-positioned by a local pin, then "5." with the four-signal graphic. Pace quickens slightly, confident.	Third, pick your battle by category, not by hype. If you're a fertility clinic or an injury firm, this isn't where your next dollar goes. If you're a med spa, a dentist, an orthodontist, or an HRT clinic in a town without big chains, the ground is open right now, and it will not stay open. Fourth, your competitor is probably a chain, not an app, and the defense is local. And fifth, there's no single lever. A real program works on all of it at once, with a scientific approach.
7:12 - 7:32	SUBSCRIBE MOMENT. Brief cut to the avatar, warm and direct, with a tasteful animated subscribe prompt in the corner (sage and antique-gold, on cool white). A small "AI Search, Studied" series badge sits beside it.	If any of this is landing, here's the one thing I'd ask. Subscribe. Next episode we take a fresh set of industries and run the same experiment, so you can watch where the ground opens up next. We do this in the open, every few weeks, and it compounds the more you follow along.
7:32 - 8:10	Annette's avatar on camera, cool-white studio, sage and forest-green accents, a framed print of the Colorado mountains under a clear blue sky behind her. Warm, direct delivery to lens. Lower-third: "Annette Thompson, SEO Believer."	A quick word on who ran this, so you know we're not guessing. Our team has been ranking websites since 1995. I personally took a nonprofit from near-zero search authority all the way to ranking nationally, on organic traffic alone, with no paid advertising. [DR STAT PENDING ANNETTE'S FRAMING DECISION] My background is in medical technology and biochemistry, and it shows in how we work. We form a hypothesis, query the AIs directly, capture the raw answers, and report what actually happened. Not what the trend pieces predict.
8:10 - 8:45	Final CTA card. Cool white, antique-gold headline. Drone shot of the Front Range under a clear blue sky with white clouds, behind a semi-transparent panel. Text: "Find out who AI names in your category." URL: seobeliever.com. Sage button graphic.	The whitespace we found is real, and it's closing as more owners wake up to it. So here's the offer. We'll tell you exactly who AI search names in your category and your city. That's the first question we answer on every engagement, and the read is most valuable right now, before your competitor reads it first. That's seobeliever.com. Come find out where you stand.
8:45 - 9:00	END SCREEN. Cool-white end-screen layout sized for YouTube elements: one video thumbnail slot on the left labeled "Watch next," one subscribe button on the right. Keep it to these two elements only. Hold ~15 seconds so both elements are clickable. Soft Flatirons linework along the bottom under a clear blue sky.	If you want the deeper write-up of this study, I've left it on screen and linked below. And subscribe so episode two lands in front of you. We'll see you in the next one.

---

YOUTUBE CHAPTERS (paste into the description)

YouTube auto-creates chapters when the description contains a timestamp list that starts at 0:00 and has at least three entries, each at least 10 seconds apart. Chapters lift average watch time by roughly 11% and qualify the video for Key Moments in Google Search. Keep the labels short and benefit-led.

0:00 An AI is naming a business in your town
0:42 Finding 1: No AI agrees on a winner
1:38 Finding 2: One page feeds every AI's price answer
2:42 Finding 3: Can you win? It depends on your industry
3:48 Finding 4: Your real competitor isn't a national app
4:28 Finding 5: Every AI rewards something different
5:08 Finding 6: When Google shows an AI Overview (and when it doesn't)
5:43 What this means for your revenue
7:12 Subscribe / what's next in the series
7:32 Who ran this study
8:10 Find out who AI names in your category

Keep the first chapter generic and curiosity-driven, not "Intro." Each chapter title doubles as a search-friendly label and a re-hook in the scrubber.

---

SLIDE-BY-SLIDE DECK OUTLINE

Use this if the avatar is delivered against a slide deck rather than (or in addition to) b-roll. Each slide maps to a beat above. Palette: cool fresh white background, antique-gold headers and rules, sage and deep-forest-green accents, Colorado-mountain imagery (always under a clear blue sky with white clouds, never a white or overcast sky). No orange anywhere. Elegant, generous whitespace, one idea per slide.

Slide 1 - The hook (cold open)

• Title: "An AI is recommending a business in your town. By name."
• On screen: a single stylized AI chat bubble naming a (blurred) local business, on near-empty white.
• Visual: no chart. Just the chat bubble and a slow-blinking cursor. Sets the stakes.

Slide 2 - What we did

• Title: "We asked the AIs what your customers are asking them."
• On screen: row of six AI marks (ChatGPT, Perplexity, Claude, Grok, Gemini, Google AI Overviews); "30+ high-intent local queries / 8 verticals / Front Range / June 2026."
• Visual: simple icon row, no data yet. Antique-gold underline.

Slide 3 - Finding 1: No AI agrees on a winner

• Title: "There is no stable winner."
• On screen: the AI agreement matrix (rows = Med Spa A, B, C, D...; columns = the six AIs; filled cells scattered; no business fills a full row; Gemini's column mostly its own names; Google AI Overviews column empty for this query).
• Visual: this is the centerpiece chart. The patchiness is the message.

Slide 4 - Finding 1, the takeaway

• Title: "Fragmentation is the opportunity."
• On screen: a single empty matrix cell glowing, with a short line: "Move from named-by-one to named-by-all."
• Visual: zoomed-in detail of the matrix, one pulsing open cell.

Slide 5 - Finding 2: One source feeds every AI's cost answer

• Title: "On pricing questions, one source feeds them all."
• On screen: the "one source feeds the answer" diagram. Center node = a single pricing page ("$3,150 start to finish"). Arrows to Perplexity, Claude, Grok, and Google AI Overviews. Gemini off to the side, dashed line, label "generic ranges only, no local source." A second empty node: "the open seat."
• Visual: hub-and-spoke diagram. The empty open seat is the call to action.

Slide 6 - The vertical-openness ladder

• Title: "Can you win? It depends on your line of work."
• On screen: the vertical-openness ladder (horizontal bars, sorted most-open to most-closed, green to forest to muted red): dental implant cost (most open) > braces cost > med spa > cosmetic dentistry/veneers > plastic surgery > LASIK/cataract > fertility/IVF > personal injury law (most closed).
• Visual: the single most screenshot-worthy graphic. One glance shows where the ground is open.

Slide 7 - Finding: The telehealth myth

• Title: "We expected national telehealth brands. We found franchises."
• On screen: the six-metro HRT graphic (Boulder, Denver, Phoenix, Nashville, Tampa, Austin). Local-independent visibility vs chain dominance per metro; Nashville and Boulder at the open end, Austin and Tampa locked. Bold annotation: "Zero telehealth brands appeared in any local answer, on any of six AIs, including both Google products."
• Visual: small US map with city pins, or a six-bar comparison. Gray-X on the telehealth placeholder.

Slide 8 - Finding: Different AIs reward different signals

• Title: "There is no single 'AI optimization' lever."
• On screen: the signal-by-AI table as a graphic. Columns: Reviews & listings | Clear, well-labeled pages | Reddit & forum presence | Well-known name in Google's index. AI marks placed under the signal each rewarded (Perplexity + ChatGPT under reviews; Claude + Google AI Overviews under clear pages; Grok under Reddit; Gemini under well-known name).
• Visual: four-column signal map.

Slide 9 - Finding: AI Overviews triggers on cost, not "best of"

• Title: "Cost questions get an overview. 'Best of' questions get the map pack."
• On screen: the AI Overview trigger-rate split. Of 20 Google queries: 11 rendered an overview, 9 returned a local 3-pack. The 11 grouped (all four cost queries; fertility; HRT in five of six metros). The 9 grouped (fragmented Boulder "best of" discovery queries).
• Visual: a clean 11-vs-9 split bar, each side annotated with what triggered it.

Slide 10 - The ROI takeaway

• Title: "Five moves, ROI-first."
• On screen: a numbered, scannable list. (1) When AI leaves you out, nothing warns you. (2) The cheapest win is the pricing page. (3) Pick your battle by category, not by hype. (4) Your competitor is a chain, not an app. (5) No single lever; run a real program.
• Visual: clean numbered list on white, antique-gold numerals, sage checkmarks.

Slide 11 - Who ran this (credibility)

• Title: "A scientific approach to getting found."
• On screen: short credibility line. "Ranking websites since 1995. Took a nonprofit from near-zero authority to nationally ranking, on organic traffic alone. [DR STAT PENDING ANNETTE'S FRAMING DECISION] Background: medical technology and biochemistry."
• Visual: Annette headshot, Colorado-mountain band (clear blue sky with white clouds, never overcast), SEO Believer mark.

Slide 12 - CTA

• Title: "Find out who AI names in your category."
• On screen: "The whitespace is closing. The read is most valuable now." URL: seobeliever.com.
• Visual: Front Range drone still (clear blue sky with white clouds, never overcast) behind a translucent white panel; sage call-to-action button; antique-gold headline.

---

AI IMAGE & INFOGRAPHIC PROMPTS

Feed these to an image generator (DALL-E, Ideogram, Midjourney) or to NotebookLM for infographic generation. Shared style spec, prepend to every prompt:

Brand style: elegant, editorial, generous whitespace. Background cool fresh white (#FBFBF8 feel). Accent palette: antique gold, soft sage green, deep forest green. Colorado Rocky Mountain motifs welcome; whenever any sky is actually depicted in a mountain visual, it must be a clear blue sky with white clouds, never a white or overcast sky. Absolutely no orange, no neon, no dark backgrounds. Thin refined lines, premium feel, calm and credible, not flashy.

1. Hero / thumbnail background

Editorial hero image on a cool fresh white background. A subtle, elegant illustration of the Colorado Flatirons and Rocky Mountains rendered in thin antique-gold and sage-green linework along the lower third. If any sky is shown, render it as a clear blue sky with a few soft white clouds, never white or overcast. Upper area mostly clean white negative space for text overlay. Soft, premium, calm. No orange, no people, no text.

2. AI agreement matrix infographic

A clean editorial infographic: a grid/matrix on cool white. Left column lists anonymized business labels "Med Spa A, B, C, D, E." Top row shows six small abstract AI icons. Some cells filled with small sage-green dots, scattered, no row completely filled. One empty cell highlighted with a thin antique-gold ring and a faint glow. Minimalist, lots of whitespace, thin lines. No orange.

3. "One source feeds the answer" hub diagram

An elegant hub-and-spoke diagram on cool white. A central rounded card labeled "$3,150 pricing page" in deep forest green. Four thin antique-gold arrows radiate outward to four small abstract AI icons. One additional icon sits apart, connected by a dashed gray line, labeled "generic only." Beside the center, a faint empty dashed card labeled "the open seat." Refined, minimal, premium infographic style. No orange.

4. Vertical-openness ladder

A horizontal bar-chart infographic on cool fresh white, premium editorial style. Eight stacked horizontal bars, each labeled with a business category. Bars are color-graded top to bottom from soft sage green (most open) through deep forest green to a muted dusty-rose red (most closed). Sorted longest/greenest at top. Thin, elegant, clearly readable labels. Calm and authoritative. No orange, no neon.

5. Six-metro HRT map graphic

A minimalist US map illustration on cool white, with six small city pins labeled Boulder, Denver, Phoenix, Nashville, Tampa, Austin. Pins in sage and deep forest green; two pins (Austin, Tampa) marked with a small "locked" padlock motif, two pins (Boulder, Nashville) glowing as "open." A small national-brand icon shown crossed out with a soft gray X. Elegant, thin linework, lots of whitespace. No orange.

6. Signal-by-AI table graphic

A clean four-column comparison infographic on cool white. Column headers: "Reviews & listings," "Clear, well-labeled pages," "Reddit & forum presence," "Well-known in Google's index." Small abstract AI icons placed under the relevant columns. Thin antique-gold dividers between columns, sage-green header band. Editorial, premium, highly legible, generous spacing. No orange.

7. AI Overview trigger-rate split

A simple, elegant split-bar infographic on cool fresh white. One large bar divided into two segments: a larger sage-green segment labeled "11 of 20: AI Overview rendered (cost questions)" and a smaller forest-green segment labeled "9 of 20: local map pack (best-of questions)." Clean numerals in antique gold. Minimal, premium, lots of whitespace. No orange.

8. Credibility / closing card background

A calm editorial background on cool fresh white. A thin band of antique-gold and sage Colorado mountain linework across the lower portion. If any sky is shown, render it as a clear blue sky with a few soft white clouds, never white or overcast. Upper two-thirds clean white for text. Premium, trustworthy, understated. No orange, no people, no text.

---

TITLE, DESCRIPTION, TAGS, THUMBNAIL

Packaging note: in 2026 YouTube weighs "Quality CTR." A clickbait title that wins the click but loses the viewer in the first 15 to 30 seconds gets demoted. Every title below promises exactly what the first 15 seconds delivers. For a brand-new channel the first subscribers come from search, so the primary title front-loads the phrase an owner would actually type ("AI search," "recommend," local-business intent) and pairs it with tension.

Primary video title (lead with this):

What AI Actually Recommends When Customers Search For A Business Like Yours

Alternate titles for A/B testing (test one variable at a time using YouTube's Test & Compare; keep them honest):

• AI Is Already Recommending Businesses By Name. Is It Yours?
• We Asked 6 AIs Who To Hire. They All Disagreed.
• Who AI Search Actually Cites: What We Found Across 8 Local Industries
• One Web Page Is Feeding Every AI's Price Answer. Here's How To Own It.

Title craft: keep the primary under ~60 characters where possible so it isn't truncated on mobile and in search. Put the most important words first. Title and thumbnail should each say something the other doesn't, and together make a promise the first 15 seconds keeps.

Description (first two lines are the only ones most viewers see before "Show More," so they carry the outcome and the primary keyword; full text aims for 150 to 300 words, links high up):

Find out what AI search actually recommends when a customer asks ChatGPT, Perplexity, Gemini, or Google to name a business like yours. We ran the experiment across 8 local industries and recorded who got named and who got left out.

🔗 See who AI names in your category and city: https://seobeliever.com 📄 Read the full written study: https://seobeliever.com/insights (update to the exact article URL)

Episode 1 of AI Search, Studied, an ongoing series from SEO Believer where our team runs real experiments on what the AIs recommend, then reports exactly what happened.

In this episode we asked six AIs (ChatGPT, Perplexity, Claude, Grok, Gemini, and Google AI Overviews) the exact high-intent questions local customers are already asking them, across eight kinds of service business on the Colorado Front Range. What we found:

• There's no stable winner. Six AIs, six different "best of" answers.
• On pricing questions, a single well-built web page is feeding nearly every AI's answer.
• Whether you can win depends on your category. Dentists and orthodontists: open. Fertility and injury law: locked.
• The telehealth fear is unfounded. Your real competitor is the multi-location franchise.
• Different AIs reward different signals, so there's no single optimization lever.
• Google's AI Overview shows up for cost questions, not "best of" questions.

If you run a service business in Boulder, Denver, or any comparable metro, this is your market, your category, and the actual answer an AI gives today. Subscribe to follow the series, because we run a fresh set of industries every few weeks.

SEO Believer has been ranking websites since 1995, with a scientific approach: form a hypothesis, query the AIs directly, capture the raw answers, and report what actually happened.

⏱️ Chapters 0:00 An AI is naming a business in your town 0:42 Finding 1: No AI agrees on a winner 1:38 Finding 2: One page feeds every AI's price answer 2:42 Finding 3: Can you win? It depends on your industry 3:48 Finding 4: Your real competitor isn't a national app 4:28 Finding 5: Every AI rewards something different 5:08 Finding 6: When Google shows an AI Overview (and when it doesn't) 5:43 What this means for your revenue 7:12 Subscribe / what's next in the series 7:32 Who ran this study 8:10 Find out who AI names in your category

#AISearch #GEO #LocalSEO #GenerativeEngineOptimization #BoulderSEO

Hashtag rule applied: 5 highly relevant tags only. The first three (#AISearch #GEO #LocalSEO) render above the title on mobile. Long hashtag lists or off-topic tags trigger spam classification, so we keep it tight and on-topic.

Tags (semantic, not stuffed; branded + 2 to 3 exact-match + a handful of long-tail):

SEO Believer, AI Search Studied, AI search optimization, generative engine optimization, GEO, local SEO, how AI search picks businesses, ChatGPT local recommendations, Perplexity local SEO, Google AI Overviews, AI search for local business, Boulder SEO, Front Range marketing

Thumbnail direction (drives CTR, the single biggest growth lever):

• Expressive human faces lift CTR meaningfully, so use Annette's avatar with a clear, curious-but-confident expression on the left third.
• High contrast, clean cool-white background so it pops against YouTube's red/white/black UI. Avoid clutter.
• 3 to 5 words of bold, thick text max. Test text-vs-no-text and face-vs-no-face as separate experiments.

Thumbnail-text options (short, high-contrast, antique-gold or deep-forest-green text on cool white):

1. IS IT YOUR NAME?
2. 6 AIs. 6 ANSWERS.
3. ONE PAGE WINS AI
4. CAN YOU WIN AI SEARCH?
5. AI NAMES BUSINESSES NOW

Thumbnail image prompt (feed to an image generator, then drop the text on in Canva):

Editorial YouTube thumbnail, cool fresh white background. On the left third, a calm, confident professional woman (Annette's avatar likeness) with a curious, slightly raised-eyebrow expression, well lit, sharp. On the right two-thirds, clean negative space for bold text overlay, with a faint stylized AI chat bubble naming a blurred business. Thin antique-gold and sage-green accents; a hint of Colorado Flatirons linework along the very bottom (if any sky shows, clear blue with white clouds). High contrast, premium, uncluttered. No orange, no neon, no dark background.

---

VOICEOVER WORD COUNT

Spoken voiceover only (the right-hand column), not counting visual directions, slide outline, prompts, or metadata: approximately 1,400 words, which reads to roughly 8.5 to 9.5 minutes at a natural avatar pace. That lands squarely in the 8-to-10-minute target, which is the data-backed sweet spot for educational and authority content (see youtube-channel-strategy.md). The added ~260 words over the prior draft are not filler; they are the retention scaffolding the research calls for: a clearer promise at 0:08, a one-line micro-hook opening each finding so the retention curve re-lifts at every section boundary, a mid-roll open loop at 3:33, an explicit subscribe moment at 7:12, and an Episode 2 tease at the close.

If a future episode needs to come in tighter, do not cut a finding. Trim the second halves of the lower-stakes findings (5 and 6) and tighten the ROI recap, and let the retention curve in YouTube Studio decide the ideal length for episodes 2 and beyond (run the 3-Length Test described in the strategy doc).

---

SHORTS CUT LIST (top-of-funnel, optional but recommended)

The strategy doc's fastest 2026 subscriber path is long-form plus Shorts. Cut 2 to 3 vertical Shorts (15 to 30 seconds each) from the most surprising beats, each ending on a hook to the full episode:

1. "We expected national telehealth brands. We were wrong." (Finding 4), the most counterintuitive result.
2. "One web page is feeding every AI's price answer." (Finding 2), the cheapest-win hook.
3. "Six AIs. One question. Six different businesses." (Finding 1), the fragmentation reveal.

Each Short: open on the claim in the first 2 seconds, deliver the one-line payoff, end with "full study on the channel, episode one." Keep the same cool-white, sage, antique-gold look; any Colorado mountain visual gets a clear blue sky with white clouds.

YouTube Channel-Growth Strategy: SEO Believer

Purpose: Ground the SEO Believer YouTube channel in current (2026) growth best practices, so the first study video is built to GROW the channel, not just narrate an article. This doc drives the revision of youtube-video-script.md. Owner: Annette Thompson, SEO Believer Date: June 2026 Voice rules carried throughout: no em-dashes, no staccato fragments, contractions, "AI" not "engine," "scientific approach" not "lab," company "we/our team" voice, do not stress the word "believers."

---

TL;DR: the recommendation for Episode 1

Recommended length: 8 to 10 minutes (target 9). This is the data-backed sweet spot for educational and authority content, and it gives the study room to breathe without padding. The current script runs about 1,140 spoken words (roughly 7 to 7.5 minutes). We are NOT cutting it down. We are letting it grow slightly toward 9 minutes by adding a stronger hook, two retention re-hooks, an explicit subscribe moment, a chapter structure, and a forward tease to Episode 2. Every added second must carry information or tension, never filler.

Recommended structure:

1. 0:00 to 0:08, cold-open hook. A bold, concrete, stakes-first claim. No logo, no "welcome to the channel." (Already strong in the draft; we sharpen it.)
2. 0:08 to 0:25, promise plus credibility flash plus first subscribe seed. Tell them exactly what they will know by the end, name the series, and plant the channel identity.
3. 0:25 onward, deliver the six findings, each as its own chapter, each opening with a micro-hook so the retention curve re-lifts at every section boundary.
4. Mid-roll re-hook (~halfway): an open loop that promises the most surprising finding is still coming.
5. ROI payoff section: the "so what for your revenue" turn.
6. Credibility close + CTA + subscribe + end-screen + Episode 2 tease.

Why this specific shape: YouTube rewards watch time and satisfaction, not raw length. A tight 9-minute video that holds 40 percent-plus retention beats a padded 15-minute one that holds 20 percent. Educational/how-to is the single highest-retention niche on the platform, and 8-to-12 minutes is its proven band. We sit at the lower-confident end of that band because this is a brand-new channel that has not yet earned long-watch trust.

---

PART 1: Optimal video length and the retention-vs-length data

The core principle

YouTube's algorithm optimizes for two things above all: how long viewers stay on the platform and how satisfied they are. Length is a lever on watch time, not a goal in itself. Padding a video to hit a runtime target reliably destroys performance.

• A 6-minute video at 80 percent retention produces 4.8 minutes of watch time per viewer. A 20-minute video at 30 percent retention produces 6 minutes on paper, but YouTube also weighs retention percentage, CTR, and whether the viewer keeps watching afterward, and consistently demotes videos people bail out of early. (YTShark, Ideal YouTube Video Length in 2026)
• The average YouTube video retains just 23.7 percent of its viewers, and 55 percent of viewers drop off within the first 60 seconds regardless of length. Only 16 percent reach the final 10 seconds of an average video. (2025 State of YouTube Audience Retention Benchmark, via YTShark)
• Real example from the source: a finance-education client posting 25-minute deep dives sat at 19 percent retention. Trimmed to 10-to-12 minutes by cutting repetition, retention climbed to 38 percent and impressions nearly doubled. (YTShark)

Length by content type (2026 data-backed bands)

Content type	Ideal length	Why
Tutorials / How-to	8 to 12 min	Enough depth without filler
Finance / Education	7 to 15 min	Complex topics need space
Entertainment / Vlogs	5 to 10 min	Highest retention sweet spot
Product reviews	8 to 15 min	Viewers want thorough comparisons
YouTube Shorts	15 to 30 sec	Completion rate drives the Shorts algorithm

(Source: YTShark, Ideal YouTube Video Length in 2026)

Cross-confirming benchmarks:

• Best-performing long-form band overall is 5 to 10 minutes; educational how-to leads all niches at roughly 42 percent average retention. (Whitehat SEO / Humble & Brag benchmark roundups, 2026)
• For B2B specifically, 5 to 10 minutes is the recommended band: enough depth for complex topics, higher watch time, more comments and shares than shorter formats. Education content should target 40 to 60 percent average view duration. (Whitehat SEO, Video Marketing Benchmarks 2026)

What this means for a NEW channel

New channels benefit from tighter, more frequent videos and faster iteration. The source advice for sub-1,000-subscriber channels: "you learn faster by publishing a 6-minute video twice a week than a 15-minute video every other week." (YTShark)

Decision for Episode 1: land at 8 to 10 minutes. It is long enough to be the definitive answer to "who does AI actually cite in my industry" (a searchable, authority-building question) and short enough that a brand-new channel can plausibly hold retention across it. After we have retention data on 3 to 5 videos, run the 3-Length Test (publish near-identical topics at ~6, ~10, ~15 min, compare retention curves after 2 weeks, keep the winner). (YTShark)

---

PART 2: Hook and retention structure that works now

The first 30 seconds is disproportionately important

A video that retains 70 percent of viewers past 30 seconds gets recommended to far more new audiences than one that loses 40 percent in the same window, even at identical total views. If you are below 50 percent retention in the first 10 to 15 seconds, your hook is not working. (Tube Analytics; SocialRails; 1of10)

The recommended hook framework (0:00 to 0:30)

• 0:00 to 0:05, attention grab: shock, tease, question, or a striking clip. Cut the intro entirely, or keep it under 5 seconds.
• 0:05 to 0:15, clarify the promise: state concretely what the video delivers. Strong promises are specific ("Master the J-cut in 30 seconds"), telling the viewer exactly what they will be able to do and on what timeframe.
• 0:15 to 0:30, establish stakes and context, start the journey. (Source: Tube Analytics, YouTube Hook Examples 2026; 1of10)

Most effective hook types for educational content

Direct-promise hooks and question hooks work best for educational content. (Entertainment thrives on pattern interrupts; behind-the-scenes thrives on curiosity gaps.) The three highest-performing hook structures overall are the bold-claim hook, the question hook, and the result-first hook. (Tube Analytics)

Re-hooks across the body (the part most creators skip)

Retention is not a single event at 0:30. Reading the retention curve in YouTube Studio:

• Sharp early drop (first 30 sec): the hook failed, not the length.
• Gradual middle decline: normal. If the curve flattens, content is holding well.
• Steep drop at a specific point: that is where the video should have ended. (Source: YTShark)

Practical move: open every chapter with a one-line micro-hook so the curve re-lifts at each section boundary, and plant one mid-roll open loop ("the finding that surprised us most is still coming"). Improving average retention by just 10 percentage points correlates with a 25 percent-plus increase in algorithmic impressions. (YTShark)

---

PART 3: Title and thumbnail packaging for click-through

CTR benchmarks (2026)

• Below 3 percent needs work, 4 to 5 percent is solid, above 7 percent is exceptional.
• By source: Search CTR is highest, 8 to 15 percent for well-optimized content; Browse/suggested CTR is 3 to 7 percent. (Humble & Brag; ThumbMagic)

The 2026 game-changer: "Quality CTR"

YouTube now actively demotes videos that earn a high click-through rate but then lose viewers in the first 15 to 30 seconds. A clear, honest title plus an accurate thumbnail earns a lower initial CTR but a much higher quality CTR, and the algorithm rewards quality CTR with sustained distribution. Translation: do not clickbait. Promise exactly what the first 15 seconds delivers. (Ampifire, Best YouTube Thumbnail Guide 2026)

Thumbnail rules

• Expressive human faces can lift CTR 20 to 30 percent (VidIQ research, via NearStream). Annette's avatar face with a clear expression is a real asset here.
• Contrast beats color choice. Bright subject on a dark or clean background wins; bright subject on a busy background fails. To stand out, avoid YouTube's own red/white/black palette dominance. (Backlinko; ThumbMagic)
• 3 to 5 words of text max. Thick, clean, readable-at-small-size fonts. One idea, not three. (Awisee; ViewsMax)
• 90 percent of the best-performing videos use custom thumbnails. (YouTube, via Backlinko)

Title + thumbnail as a pair

The title and thumbnail should each say something the other does not, together forming a promise the video keeps within the first 15 seconds. Optimized titles lift CTR 25 to 40 percent. For a new channel, the first 1,000 subscribers come from search, so the title should contain the exact phrase people type. (Teleprompter; Influencer Marketing Hub via Awisee; YTShark)

A/B testing

Test one variable per experiment (text vs no text, face vs no face, color A vs B). YouTube's native thumbnail Test & Compare can run this. (ThumbMagic; Spotter Studio)

---

PART 4: Description optimization

First two lines are prime real estate

The first 2 to 3 lines show before "Show More" and matter most for both search and AI systems. Do not waste them on "Welcome back to my channel." State the outcome and the primary keyword immediately, in the first sentence. (InfluenceFlow; Teleprompter)

Keywords: semantic, not stuffed

Creators using natural keyword integration saw 23 percent higher impressions than those using outdated keyword-stuffing tactics (Influencer Marketing Hub 2026, via search). Cover the topic thoroughly rather than chasing a density number. Primary keyword in the first sentence. (InfluenceFlow)

Length and structure

• Up to 5,000 characters available. Descriptions of 150 to 300 words rank better than short ones; aim for 200-plus. Short paragraphs of 2 to 3 sentences. (Touhfa; VidIQ)
• Include the key links high up: the site URL (seobeliever.com), and the companion article. Links in the description are a primary discovery and conversion path.

---

PART 5: Tags, hashtags, chapters, cards, end screens, playlists

Tags

Still useful, no longer the primary lever. Use: branded tags (channel name, series name), 2 to 3 exact-match primary-keyword tags, 3 to 5 long-tail phrases from the content. (InfluenceFlow)

Hashtags

3 to 5 highly relevant hashtags maximum. Long lists trigger spam classification; broad or off-topic tags (e.g. #Gaming on a tax video) damage ranking because the algorithm scores hashtag relevance against actual content. (InfluenceFlow)

Chapters / timestamps

Videos with chapter markers see roughly 11 percent higher average watch time because viewers jump to relevant sections. Timestamps also qualify the video for Key Moments segments in Google Search, pulling in off-platform traffic. Essential for anything past a few minutes, especially structured/educational content. (InfluenceFlow; YouTube 2026 creator data via search)

Cards

Place cards at the 50 to 75 percent mark, not early (do not interrupt the hook) and not so late the viewer has dropped. Cards plus end screens lift session time 20 to 35 percent when well placed. (Tube Analytics)

End screens

• Best configuration for a business channel: two elements only. One video card to the most relevant next video, plus one subscribe button. More elements split attention and reduce clicks on the primary card.
• Duration 4 to 20 seconds (10 and 15 are the most-tested); design the final 15 to 20 seconds of the video to host them.
• A click from an end-screen card counts as a session continuation, one of the strongest positive algorithm signals. (Gyre; Tube Analytics; Backlinko)

Playlists

• Playlists increase Session Time, which YouTube explicitly rewards ("your channel gets a boost when people watch anything anywhere on YouTube after your content"). (Backlinko, quoting YouTube Creator Academy)
• The first 3 videos in a playlist decide whether viewers continue. Playlist links in end screens add ~3.2 extra minutes of watch time per session. The 45-minute-to-2-hour total-length band performs best. (YouTube Playlist Analyzer; Tube Analytics)
• Action: create a playlist now (e.g. "AI Search, Studied") and make this study video its first entry, so every future episode compounds session time.

---

PART 6: Posting cadence for a new channel

• Consistency beats frequency. Channels with a regular, predictable schedule grow faster than channels that post more but erratically. (Navigate Video; John Isaacson)
• Minimum for algorithm momentum: 1 long-form video per week. Posting less than weekly makes it hard to build algorithm trust because the model of the channel degrades between uploads. Most successful creators publish 1 to 3 high-quality videos per week. (Veefly; thisbahamiangyal)
• Sustainability is the real constraint: "one excellent video every week for a year outperforms three mediocre videos per week for three months followed by silence." (John Isaacson)
• New-channel nuance: shorter, more frequent videos accelerate learning and iteration early on. (YTShark)

Recommendation for SEO Believer: commit to a cadence Annette can actually sustain. A realistic, defensible target is one substantive long-form episode every 1 to 2 weeks, optionally supported by Shorts cut from each episode for top-of-funnel discovery. Do not promise weekly if it cannot be kept; predictability matters more than the number.

---

PART 7: What specifically helps a brand-new channel gain subscribers and authority

1. Win search first. The first 1,000 subscribers come from search, not the recommendation algorithm. Make searchable content, put the exact search phrase in the title, and show the answer or solution within the first 30 seconds. (YTShark; Newzenler) This study is perfectly shaped for it: "who AI search actually cites in [your industry]" is a real, high-intent query.
2. Pair Shorts with long-form. Shorts bring fast exposure and introduce the voice in under a minute; long-form builds trust. Combining them is the fastest 2026 subscriber path. Cut 2 to 3 Shorts from each episode's most surprising findings. (vidIQ; Newzenler)
3. Niche clarity + consistent packaging. A clear niche, consistent title/thumbnail style, and a named recurring series tell both viewers and the algorithm what the channel is. Frame the channel around an ongoing research series. (TubeBuddy; ImpactPlus)
4. Demonstrate E-E-A-T on camera. Experience, expertise, authoritativeness, trustworthiness: state credentials and real results plainly (years of experience, the nonprofit climb, the medical-technology background). This is exactly what AI systems and viewers both reward, and it doubles as GEO signal. (Siege Media, How to Rank in AI Search)
5. Momentum takes volume. Most channels see real traction after 20 to 50 videos. Treat Episode 1 as the start of a compounding library, not a one-off. (Newzenler)
6. Drive session continuation. Every subscribe button, end-screen card, and playlist placement that keeps a viewer on YouTube after this video is a direct, compounding ranking signal. Build them in from Episode 1. (Backlinko; Tube Analytics)

---

Sources

• Ideal YouTube Video Length in 2026: What the Data Actually Says (YTShark)
• How to Grow YouTube Channel: Expert Tips (Backlinko)
• Video Marketing Benchmarks 2026 (Whitehat SEO)
• YouTube Audience Retention Benchmarks 2026 (Humble & Brag)
• YouTube Hook Examples 2026: Keep Viewers Past 30 Seconds (Tube Analytics)
• How to Hook Viewers in the First 30 Seconds (1of10)
• YouTube Audience Retention 2026 (SocialRails)
• Best YouTube Thumbnail Guide & Best Practices 2026 (Ampifire)
• YouTube Thumbnail Best Practices & Statistics (Awisee)
• YouTube CTR Benchmarks 2026 (ThumbMagic)
• Optimized Video Titles and Descriptions 2026 (InfluenceFlow)
• YouTube Metadata and Descriptions Guide 2026 (InfluenceFlow)
• How to Write a YouTube Title and Description 2026 (Teleprompter)
• YouTube Description Best Practices 2026 (Touhfa)
• How to Create High-Converting YouTube End Screens (Gyre)
• YouTube Cards and End Screens Checklist 2026 (Tube Analytics)
• Best Playlist Length for Watch Time in 2026 (YouTube Playlist Analyzer)
• How Often Should Brands Post on YouTube 2026 (Navigate Video)
• How Often Should You Post on YouTube: Business Channels 2026 (John Isaacson)
• YouTube Posting Strategy 2026 (Veefly)
• New YouTube Channel in 2026: How to Start, Grow, Get Seen (TubeBuddy)
• How to Grow a YouTube Channel in 2026 for Creators (Newzenler)
• How to Grow Your YouTube Channel in 2026 (vidIQ)

SEO Believer X (Twitter) Launch Kit

Created: 2026-06-02 For: Annette Thompson, SEO Believer Purpose: Everything needed to launch and grow an X account for SEO Believer the right way: build a genuinely helpful presence first, earn followers, then drop the original research and promote the study.

---

READ THIS FIRST: nothing here is posted yet

This is a prep-and-drafts kit. Posting requires an X account for SEO Believer that does not exist yet. Nothing in this file has been published. Once Annette creates the account and hands over access (or posts manually), these drafts are ready to go in the sequence described.

The play, in order:

1. Set up the account: avatar, header, bio, handle.
2. Follow the right people (Section 4) and start showing up helpfully.
3. Post value-first content (Section 5) for several weeks. No selling. No study yet.
4. Build relationships by reposting and quoting the right accounts (Section 6).
5. Once there's a real, engaged following, drop the "how I used Claude Code" thread (Section 7) to earn reach from the AI-coding crowd.
6. Then, and only then, post the research-promo thread (Section 8) so the study lands on warm ears, not an empty room.

A note on voice, applied to every word below: no em-dashes, no staccato fragments, contractions throughout, plain-spoken, company/team voice ("we" and "our team"), say "AI" and "the AIs" rather than "an engine," and the word "believers" never gets stressed. The brand is "Search Everywhere Optimization": SEO across Google, Maps, and the AIs, measured by ROI rather than clicks.

---

1. Logo direction

Concept rationale

SEO Believer's whole pitch is a calm, science-literate consultancy that happens to sit at the foot of the Rockies and gets businesses found everywhere their customers look. The logo has one job: say "credible, optimistic, Colorado, growth" in a mark that still reads at 32 by 32 pixels as an X avatar.

The motif that does this is the one already developed across the brand: a single antique gold line that rises like a growth curve and resolves, at its peak, into a Colorado mountain summit, set against a clear blue sky with white clouds on a clean white field. The gold line is the data and the ROI. The mountain is the climb and the place. The blue sky is the optimism, and it is non-negotiable that the sky is a confident daytime blue, never white or overcast. That single idea (a chart that becomes a mountain) carries the brand and gives us a mark that is distinctive without leaning on tired SEO chart icons or generic tech tropes.

For X specifically, the mark must survive a tiny circular crop. So the avatar version is the pictorial mark alone (gold line into a peak, sliver of blue sky), tightly composed inside a circle, with the wordmark reserved for the header where there's room to breathe.

The final logo will be generated from these prompts and then refined by hand (vector cleanup, exact gold value #B5893C, kerning on the wordmark). Treat these as starting prompts, not the finished file.

Logo prompts (copy-paste ready)

Prompt 1: Primary pictorial mark (the chart-becomes-mountain), tight circle-safe composition

A minimalist logo mark for a premium consultancy called SEO Believer, designed to read clearly at very small sizes. Cool fresh white background. Centered: a single elegant thin antique gold line (hex #B5893C) that rises from the lower left like a growth curve and resolves at its highest point into a clean, iconic Colorado mountain summit. One smaller sage-green peak sits beside it, and the nearest peak base grounds in deep forest green. Behind the peaks, a small sliver of clear blue daytime sky, the blue deeper at the top and lighter near the horizon, with one tiny soft white cloud. The composition is tight, balanced, and instantly legible. Flat vector style, crisp clean edges, thin confident strokes, generous negative space, premium and understated.

Negative: no orange, no sunset, no golden hour, no warm haze; no white, gray, hazy, or overcast sky; no dark or black background; no clutter, no fine detail that vanishes when small; no muddy gradients; no circuit boards, no glowing tech, no robots, no generic SEO chart-and-magnifying-glass cliches. Must stay legible at 32 by 32 pixels.

Prompt 2: Horizontal lockup (mark plus wordmark) for the X header and site

A horizontal logo lockup for a consultancy called SEO Believer. Cool fresh white background, bright and airy with generous negative space. On the left, a compact pictorial mark: a single thin antique gold line (hex #B5893C) rising like a growth curve and resolving into a Colorado mountain summit, a smaller sage-green peak beside it, deep forest green at the nearest base, and behind the peaks a slim band of clear blue sky with one or two small soft white clouds. To the right of the mark, the wordmark "SEO Believer" set in a refined elegant sans serif, deep forest green, with comfortable letter spacing. Optional small tagline beneath in a lighter weight: "Search Everywhere Optimization." Clean, calm, credible, premium. Flat vector aesthetic, crisp edges, subtle soft shadow under the nearest ridge only.

Negative: no orange, no sunset tones, no warm haze; no white, gray, hazy, or overcast sky; no dark background; no clutter; no lens flare or glow; no tech or robot iconography; no cheesy stock imagery. Keep the wordmark crisp and perfectly legible.

Prompt 3: Monogram fallback ("SB") for ultra-small placements and favicon

A simple circular monogram logo for SEO Believer. Cool fresh white circular field. Centered: the letters "SB" in a refined elegant sans serif, deep forest green, with a single thin antique gold line (hex #B5893C) sweeping up beneath or through the letters and tipping into a tiny mountain peak at its top right, suggesting growth and a summit. A faint sliver of clear blue sky with one tiny white cloud sits behind the peak. Balanced, iconic, and legible at favicon size. Flat minimal vector style, crisp edges, thin clean strokes.

Negative: no orange, no sunset, no warm tones beyond the gold line; no white, gray, hazy, or overcast sky; no dark or black field; no clutter, no detail that disappears when tiny; no gradients that muddy the mark; no tech or robot iconography. Must read clearly at 16 by 16 pixels.

Prompt 4: Pure abstract option (no mountain, for a more corporate read)

An abstract minimalist logo mark for a consultancy called SEO Believer, conveying upward growth measured everywhere. Cool fresh white background. Centered: a single elegant antique gold line (hex #B5893C) that rises in a confident upward arc and, at its peak, branches into three short thin rays pointing to three small dots, suggesting one source being found across multiple places. The arc grounds at the lower left in a small deep forest green anchor shape. A faint sage-green secondary arc echoes behind it for depth. Clean, premium, geometric, calm. Flat vector style, crisp edges, thin strokes, generous negative space.

Negative: no orange, no sunset, no warm haze; no dark or black background; no clutter; no glowing tech, no circuit boards, no robots; no magnifying glass; no generic chart-junk. Must read clearly at 32 by 32 pixels.

Recommendation: Prompt 1 is the lead candidate for the X avatar because the chart-becomes-mountain idea is the brand's signature and it stays legible tiny. Prompt 2 is for the header. Generate several, pick the cleanest, then refine the vector by hand and lock the exact #B5893C gold.

---

2. X profile graphics

Avatar (400 x 400, displays as a circle)

Use the finished logo mark from Section 1 (Prompt 1 output). If generating fresh art rather than dropping in the refined logo file, use this:

A circular profile avatar for SEO Believer, 400 by 400 pixels, designed to read clearly inside X's circular crop. Cool fresh white field filling the circle. Centered and tightly composed: a single thin antique gold line (hex #B5893C) rising like a growth curve and resolving into a clean Colorado mountain summit, one smaller sage-green peak beside it, deep forest green at the nearest base, and a small sliver of clear blue daytime sky behind the peak with one tiny soft white cloud, the blue deeper at top and lighter toward the horizon. Simple, balanced, premium, instantly legible at small size. Flat vector style, crisp edges, thin clean strokes.

Negative: no orange, no sunset, no warm haze; no white, gray, hazy, or overcast sky; no dark or black field; no clutter, no detail that vanishes when small; no tech or robot iconography; no magnifying glass. Keep the key shapes inside the central circle so nothing important gets cropped. Must read at 32 by 32 pixels.

Note: keep all critical detail well inside the center. X crops the avatar to a circle and shrinks it hard in feeds.

Header / banner (1500 x 500)

A wide, elegant X (Twitter) header banner for SEO Believer, exactly 1500 by 500 pixels. Cool fresh white background as the dominant base, bright and airy with generous negative space. Across the lower two-thirds, a refined range of Colorado mountains climbing gently from lower left to upper right, drawn in clean flat layered shapes: nearest ridgelines in deep forest green, mid-distance hills in sage green, receding ranges in progressively lighter sage for depth. A single confident antique gold line (hex #B5893C) traces the top ridgeline like a growth curve that has become a mountain skyline, peaking toward the right. Above the peaks, a clear blue daytime sky grading from a deeper blue at the top to a lighter blue at the horizon, with three or four soft rounded white cumulus clouds at varied heights. Keep the right third and the lower-left corner relatively clean, because X overlays the circular avatar at the lower left and a profile card at the bottom. Optional small wordmark "SEO Believer" or tagline "Search Everywhere Optimization. Google, Maps, and the AIs." set in a refined sans serif in deep forest green, placed in the upper-right clean space. Calm, optimistic, premium, science-literate. Flat editorial illustration aesthetic, crisp edges, subtle atmospheric depth between mountain layers.

Negative: no orange, no sunset, no golden hour, no warm wash; no white, gray, hazy, washed-out, or overcast sky; no dark or black areas; no clutter, no busy foreground detail, no lens flare; no circuit boards, no glowing tech, no robots; no cheesy stock photography. The sky must be clearly, confidently blue with white clouds.

Note on safe zones: X covers the lower-left corner of the banner with the round avatar and renders profile text along the bottom. Keep the avatar art and any header text out of the lower-left corner and away from the very bottom edge. On mobile, X also crops the top and bottom of the banner slightly, so keep the key mountain ridge and any wordmark vertically centered, not jammed against the top or bottom.

---

3. X bio

X bios cap at 160 characters. Both options below are under that limit. Pick one, swap the phone or link as needed.

Primary bio (153 characters):

Search Everywhere Optimization: we get you found on Google, Maps, and the AIs. ROI over clicks. HRT/TRT, dental, med spas. Boulder, serving nationwide.

Slightly longer alt (for the pinned tweet or "about" framing, not the 160-char bio field, since it runs long):

We're SEO Believer. Search Everywhere Optimization, which means getting you found across Google, Maps, and the AIs like ChatGPT, Perplexity, and Gemini, not just one of them. We measure in revenue, not clicks. Founder Annette Thompson has been ranking sites since 1995 and brings a medical-technology and biochemistry background to a scientific way of working: form a hypothesis, query the AIs directly, capture what actually happened, report it honestly. Flagship specialty is HRT and TRT clinics, plus dentists, orthodontists, med spas, and other local service businesses. Based in Boulder, working nationwide.

Profile field tips: Location field = "Boulder, CO." Website field = seobeliever.com. Use the longer alt as the first pinned tweet so visitors get the full story without cramming it into 160 characters.

---

4. Who to follow on X

Follow these before posting much. It seeds your timeline with the right material to reply to, quote, and learn from, and a thoughtful following list signals what the account is about. Where a handle is uncertain, the person or account is described clearly so it's easy to confirm before following. Verify each handle on X before following, since handles change.

SEO and GEO thought leaders (non-competing, national/global)

• @aleyda: Aleyda Solis, independent SEO consultant, runs SEOFOMO newsletter and Learning SEO. One of the most respected voices on SEO and increasingly on AI search. Not a local competitor.
• @lilyraygo: Lily Ray, SEO and SGE/AI Overviews researcher. Posts heavily on how Google's AI surfaces sources. Core GEO follow.
• @rustybrick: Barry Schwartz, Search Engine Roundtable. Breaks Google and AI-search news daily. Essential for staying current.
• @JohnMu: John Mueller, Google Search Advocate. The closest thing to an official Google search signal source.
• @dr_pete: Dr. Pete Meyers, longtime SEO scientist (Moz background). Data-driven, science-literate, matches the brand's tone.
• @randfish: Rand Fishkin, founded Moz, now SparkToro. Audience-research and visibility thinking beyond pure rankings.
• @MarieHaynes: Marie Haynes, focuses on Google quality, E-E-A-T, and AI in search. Strong on the trust signals that feed AI citation.
• @methode: Fili Wiese, technical SEO, ex-Google. Good for the technical-credibility side.
• @JuliaMcCoy: Julia McCoy, writes prolifically on AI and content/search. Active on the GEO conversation.
• @iPullRank: Mike King, iPullRank, deep technical and AI-search thinking. Publishes serious GEO research.

Local SEO voices

• @JoyanneHawkins: Joy Hawkins, Sterling Sky, one of the sharpest local-SEO and Google Business Profile experts. Top follow for the Maps side.
• @DarrenShaw_: Darren Shaw, founder of Whitespark, runs the annual Local Search Ranking Factors study. Essential local-SEO data source.
• @bright_local: BrightLocal, local-SEO software and original local-search research. Good repost source.
• @localseoguide: Andrew Shotland, Local SEO Guide. Veteran local practitioner.
• @Gt_Colan / @sterlingsky: Sterling Sky's agency account, posts local-SEO case studies and GBP changes (confirm exact handle).
• @StreetFightMag: Street Fight, covers local commerce and local marketing trends.

AI-search researchers and analysts

• @dmnsgn or the Perplexity official account @perplexity_ai: track how Perplexity surfaces sources.
• @OpenAI and @OpenAIDevs: for ChatGPT search and API changes that affect citation behavior.
• @GoogleSearchLiaison: Google's official search-liaison account, announces AI Overviews changes.
• @Kevin_Indig: Kevin Indig, growth and SEO strategist who publishes data-heavy pieces on AI search and Google. Research-forward, matches the brand.
• @gefen or similar AI-search analysts (describe-and-confirm): look for analysts publishing primary AI-citation studies (the small community doing what SEO Believer does). Follow the ones with real data, not hot takes.
• @dejanseo: Dan Petrovic, DEJAN, does genuinely experimental SEO and LLM research. Science-first.

Target-vertical marketing communities (dental, med spa, HRT/aesthetics)

• @DrLenTau or dental-marketing voices (describe-and-confirm): search "dental marketing" and follow the practitioners and agencies sharing real practice-growth data (for example the hosts of well-known dental-practice-growth podcasts).
• @Medspa adjacent accounts: search for med-spa and aesthetics practice-management accounts and the major aesthetics conferences. Follow the operators and consultants, not just product brands.
• American Med Spa Association (AmSpa): follow their official account for med-spa business and regulatory content; useful for understanding the flagship-adjacent vertical.
• HRT / TRT and longevity clinic voices: describe-and-confirm: follow physicians and clinic operators in hormone optimization and longevity who post about practice growth and patient acquisition. This is the flagship vertical, so build relationships here deliberately.
• @dentaltown: Dentaltown, large dental community. Good for understanding what dentists talk about.
• Modern Aesthetics / aesthetics trade media: follow the trade publications for med spas and aesthetics for repost-worthy industry content.

Claude Code and AI-coding-tool creators

• @AnthropicAI: Anthropic's official account, makers of Claude and Claude Code.
• @claudeai: the Claude product account, posts Claude Code updates and use cases.
• @alexalbert__: Alex Albert, Anthropic, posts heavily on Claude Code workflows and prompting.
• @simonw: Simon Willison, writes the most-cited independent commentary on LLM tooling and coding agents. Hugely influential, frequent reshares.
• @swyx: Shawn Wang, AI engineering community (Latent Space). Big amplifier for genuinely useful AI-builder content.
• @mckaywrigley: Mckay Wrigley, builds and demos AI-coding workflows to a large audience. High repost potential for a good Claude Code thread.
• @levelsio: Pieter Levels, indie builder, large audience that loves "I shipped this with AI" stories. Repost potential if the thread is concrete.
• @GergelyOrosz: Gergely Orosz, The Pragmatic Engineer. Reaches a serious builder audience; reshares substantive AI-tooling write-ups.
• @eric_horvitz or AI-research voices (describe-and-confirm): follow a couple of credible AI researchers for grounding, but weight the list toward builders who reshare practical workflows.
• @OfficialLoganK: Logan Kilpatrick (Google AI / Gemini developer relations). Useful for the Gemini side of GEO and for AI-builder reach.

Total: roughly 30 accounts across five buckets. Confirm each handle on X first. Lead with the non-competing SEO/GEO and AI-search people for learning and credibility, and deliberately cultivate the dental/med-spa/HRT and Claude Code buckets, because those are where reach and clients actually come from.

---

5. Ten value-first posts (build followers before the research drops)

These are fully drafted, ready to post (pending the account). Every one gives something useful and asks for nothing. No brag, no self-promo, no link to SEO Believer, no mention of the study. Space them out over several weeks, one every couple of days, mixed with replies and reposts. The goal is for a stranger in dentistry, med spa, or HRT to read one and think "these people actually know this."

Post 1: The core reframe

Search is not one place anymore. Your future customers ask Google, they ask Maps, and more of them every month ask an AI like ChatGPT or Perplexity. If you only optimize for one of those, you are invisible in the other two. We think of it as getting found everywhere, not ranking in one box.

Post 2: A genuinely useful local-SEO tip

If you run a local business and you do one thing this week, make it this: open your Google Business Profile and check that your primary category is the most specific one that fits. "Med spa" beats "spa." "Orthodontist" beats "dentist." The category is one of the strongest signals for who shows up in the map pack, and most owners set it once and never look again.

Post 3: AI search, demystified, no jargon

When someone asks ChatGPT "best dentist near me," it does not think. It reaches for the clearest, most quotable sources it can find and repeats them. So the question for your business is simple: when an AI goes looking for an answer in your category, is your information clean enough and clear enough that it gets quoted? Usually the answer is no, and that is fixable.

Post 4: The pricing-page insight (useful, not promotional)

Here is something we keep seeing. When people ask an AI what a procedure costs in their city, the AI almost always pulls the number from whoever published the clearest, most specific, dated pricing page. One good page. If your prices are buried in a PDF or a "call us" gate, you are handing that citation to the one competitor who wrote theirs down plainly.

Post 5: Reviews, framed honestly

Reviews are not vanity. For local search and for the AIs that lean on Maps, a steady stream of recent reviews is one of the few signals you fully control. Not a burst of ten in one week. A few new ones, every week, forever. The businesses that win the map pack are usually just the ones that never stopped asking.

Post 6: A "what most people get wrong" post

A common mistake: treating "show up in ChatGPT" and "show up in Google" as the same job. They are not. The AIs each reward different things. Some lean on reviews and listings. Some lean on clean, well-organized pages. One leans heavily on what people say in Reddit threads. Optimizing for "AI" as if it is one thing is how you end up optimizing for none of them.

Post 7: A free, do-it-yourself check

Try this today. Open ChatGPT, Perplexity, and Gemini, and ask each one to recommend a business in your category and city. Write down who they name. You will probably notice two things: the names are different on each AI, and you might not be on any of them. That five-minute test tells you more about your AI visibility than most paid audits.

Post 8: Local-business reassurance (the telehealth myth, generalized)

Worried a national app is going to eat your local clinic's lunch in AI search? We looked hard at this for local health and wellness, and the AIs overwhelmingly recommend local, in-person businesses for "near me" questions. The real competition is not some app. It is the multi-location chain planting flags in your map pack. That is a local fight, and a local independent can win it.

Post 9: ROI over clicks, the philosophy

Traffic is not the goal. We have seen businesses celebrate a traffic spike that brought zero new patients, and quietly ignore a small page that booked appointments every week. Clicks feel good. Revenue pays the staff. Optimize for the searches made by people who are about to spend money, and let the vanity numbers fall where they may.

Post 10: A teaching post on freshness

The AIs and Google both quietly favor pages that look current. A page dated three years ago reads as stale, even if the information is fine. One of the cheapest wins for a local business is to keep your key pages dated and lightly updated. Same content, current date, a small refresh. It is unglamorous and it works.

Cadence note: Mix these with quote-tweets and genuine replies to the accounts in Section 4. The posts give value; the replies build the relationships. Hold all study references until the account has real engagement.

---

6. Repost and quote-tweet targets

Two lists. The first is whose content to reshare or quote to build relationships and stay current. The second is who is most likely to reshare SEO Believer if the content earns it. Confirm handles before acting.

Accounts to repost or quote (to build relationships and stay current)

• @aleyda (Aleyda Solis): quote her SEO/GEO threads with a concrete local-business angle she didn't cover. High-credibility surface, and she engages with thoughtful adds.
• @lilyraygo (Lily Ray): reshare her AI Overviews findings and add what it means for a local service business specifically. Her audience is exactly who should see the brand.
• @JoyanneHawkins (Joy Hawkins): repost her Google Business Profile and local-SEO updates. The Maps community pays attention to her, and a useful add gets seen.
• @DarrenShaw_ (Whitespark): quote the Local Search Ranking Factors data with a "here's how this plays out for a med spa" translation.
• @rustybrick (Barry Schwartz): reshare his AI-search news fast, with a one-line "what this means for owners" take. Being early and useful on his news is a reliable visibility play.
• @GoogleSearchLiaison: quote official AI Overviews announcements and translate the jargon for a non-technical owner. Adds genuine value to a noisy thread.
• @Kevin_Indig: reshare his data-heavy AI-search pieces and connect them to local verticals he writes about in the abstract.
• BrightLocal / @bright_local: repost their original local-search research with credit; they tend to engage with accounts that amplify their studies well.
• Dental, med-spa, and HRT trade accounts: quote their industry posts with a search-visibility angle the trade media rarely covers. This is how the brand becomes known inside the flagship verticals.
• @simonw (Simon Willison): when he posts about how LLMs choose or cite sources, quote it and connect it to the practical "how do I get cited" question owners actually ask.

Accounts likely to reshare SEO Believer (if the content earns it)

• @simonw (Simon Willison): reshares genuinely novel, concrete LLM findings. The "how I used Claude Code to run a real study" thread is squarely his lane.
• @swyx (Shawn Wang): amplifies practical AI-engineering and AI-builder content. A clear, non-hypey workflow thread is reshare-worthy to his audience.
• @mckaywrigley: loves concrete "I built a real thing with AI" demonstrations. The Claude Code thread fits, especially with specifics on prompts and agents.
• @levelsio (Pieter Levels): his audience rewards honest "shipped this with AI, here's exactly how" stories. Keep it concrete and unhyped and it can travel.
• @alexalbert__ (Anthropic): Anthropic folks reshare strong, real-world Claude Code use cases. An original research project run via Claude Code is a credible example they like to surface.
• @claudeai / @AnthropicAI: the official accounts occasionally amplify standout community use cases. A measurable, honest study is the kind of thing that gets noticed.
• @lilyraygo (Lily Ray): likely to reshare a rigorous, honest AI-citation study because original GEO data is scarce and she values methodology.
• @DarrenShaw_ / @JoyanneHawkins: the local-SEO community reshares solid original local-search research. The anonymized study is genuinely new data for them.
• @Kevin_Indig: reshares well-built original research that fits his data-forward feed.
• Dental / med-spa / HRT trade accounts and newsletters: likely to reshare a study about their own vertical's AI visibility, because it is directly relevant to their readers and rare.

The pattern: earn reshares by being the source of something true and new (the study, the Claude Code workflow), not by asking. Give value to the first list consistently and the second list starts watching on its own.

---

7. The "how I used Claude Code" thread

Aimed at the Claude Code and AI-coding audience. The goal is reshares from builders like Simon Willison, swyx, and Mckay Wrigley, so it has to be concrete, honest, and free of hype. It tells the true story: a founder with a science background directed an AI assistant to run a real SEO/GEO research study, and here is how. Post this once the account has genuine engagement, and it can double as the bridge into the study.

Tweet 1 (hook):

I run an SEO consultancy. Last week I wanted to know who AI search actually names when people ask "best dentist in my city," so instead of guessing, we ran a real study. The interesting part for this crowd: I directed Claude Code to do most of the heavy lifting. Here's how it worked.

Tweet 2:

First, the question had to be falsifiable. My background is in medical technology and biochemistry, so I treat marketing like a lab. We wrote it as a hypothesis: AI search already names local businesses, and the named set is unstable across different AIs. Then we set out to actually test it.

Tweet 3:

I didn't ask the AI to "do SEO research." That gets you mush. I gave it a precise protocol: a fixed list of 30-plus high-intent queries, six AIs to query, the exact two query shapes to compare, and the instruction to capture answers word for word and never invent a number that wasn't in the data.

Tweet 4:

The prompts that worked were the boring, specific ones. "Capture the live output verbatim." "Record which businesses were named, in order." "Where a figure is not in the captured text, write 'not in captures,' do not estimate." Constraining what it's allowed to make up is half the work.

Tweet 5:

We leaned on a few capabilities. A research agent to run the query sweep and pull live answers. A browser tool to capture what each AI actually returned rather than what it claims it returns. And a writing pass with strict editorial rules baked in, so the draft came out in our voice instead of generic AI prose.

Tweet 6:

The honesty guardrails mattered most. I made it document every limit out loud: small sample, two cities for most verticals, a single-location IP that can skew out-of-town results, one AI answering on a lighter model when signed out. The credibility of a study is mostly in admitting what it can't tell you.

Tweet 7:

What surprised me: the assistant was great at the structured, repeatable parts and genuinely bad at judgment. It would happily over-claim if I let it. So the division of labor became clear. It runs the protocol and drafts. I decide what's true, what's defensible, and what gets cut.

Tweet 8:

Net result: a real, reproducible study with raw captures behind it, produced in days instead of weeks, by one person directing an AI assistant carefully. Not "AI wrote my research." More like a very fast, very literal research assistant that needs a strict supervisor.

Tweet 9 (close, soft bridge):

The findings themselves are genuinely surprising, and I'll share them in a separate thread soon. If you build with these tools, the takeaway is the unsexy one: the leverage is in the protocol and the guardrails, not the magic. Specific instructions and hard limits beat clever prompts every time.

Note: This thread is authentic to how the work was actually done (research agent, browser capture, strict editorial and honesty constraints). Keep it free of product names beyond Claude Code itself, and resist the urge to inflate. The builder audience rewards candor and punishes hype.

---

8. The research-promo thread

Promotes the anonymized study's real findings, drawn from research/studies/who-ai-search-actually-cites-PUBLIC.md. Post this AFTER the follower-building phase and ideally after the Claude Code thread has warmed up the audience. No business names anywhere, in keeping with the anonymized version. End with a link to the published study page once it is live.

Tweet 1 (hook):

We asked ChatGPT, Perplexity, Claude, Grok, Gemini, and Google's AI Overviews the exact questions your future customers are already asking them. "Best med spa in Boulder." "Dental implants cost." We captured every answer word for word and recorded who got named and who got left out. Here's what we found.

Tweet 2:

Finding one: AI already names businesses by name, and there is no stable winner. Ask six AIs the same "best [category] near me" question and you get a different lead name almost every time. One business named by three AIs was left out entirely by another. That instability is the whole opportunity.

Tweet 3:

Finding two, and this is the one we keep coming back to. On pricing questions, one source often feeds every AI. We asked about implant cost and four different AIs, plus Google's own AI Overview, all reached for the same single well-built pricing page. Not a directory. One clear, dated, number-dense page.

Tweet 4:

The lesson is blunt. Pricing questions reward whoever publishes the clearest dated, number-rich page that's easy for an AI to read, and most of the AIs will quote it. Right now, in several categories, that's one source. There's room for exactly one more. Being that source means getting quoted across the board at once.

Tweet 5:

Finding three: whether AI search is worth your time depends on your line of work. For dentists and orthodontists asking about cost, the win is cheap and available right now. For med spas, cosmetic dentists, and HRT clinics, the early-mover advantage is sitting there for whoever grabs it first.

Tweet 6:

For fertility clinics and personal injury firms, it's a longer, tougher climb, because entrenched regional names and heavy ad spend already hold the ground. That's not "give up." It's "start where the ground is open, build proof there, and come back to the hard categories with momentum."

Tweet 7:

Finding four, for local health and wellness especially. We expected AI to push national telehealth apps for "near me" questions. It didn't. Across six metros and every AI we tested, AI surfaced local, in-person clinics. The real competition is the multi-location chain planting flags in your map pack, not an app.

Tweet 8:

Finding five: there is no single "AI optimization" lever. Each AI rewards different things. Some lean on reviews and listings. Some reward a clean, well-organized page. One leans on genuine Reddit and forum presence. One rewards being a well-known, established name in Google's index. Anyone selling you a one-trick fix is selling you one AI's answer.

Tweet 9:

One more, from the Google data. Google's AI Overview shows up for cost questions and consolidated categories, but for genuinely fragmented "best of" questions it declines to summarize and hands you the local map pack instead. That split tells you which game you're in: be a cited source, or win the map pack.

Tweet 10 (honest close + CTA):

A note on method, because it matters. This is a small, recent, honest sample, six AIs, captured live, limits stated plainly. We withheld the business names out of professional courtesy. If you want to know who AI search names in your category and your city, that's the first question we answer on every engagement. The full study is here: [link to the published study page once live].

Notes: No business names appear, matching the public version. The "$3,150" figure from the source study was deliberately generalized to "the same single well-built pricing page" so no practice is identifiable. Replace the bracketed link in tweet 10 with the live study URL before posting. The closing line ("the whitespace closes as more businesses wake up to it") is the study's own framing and is worth echoing in a follow-up reply if engagement is strong.

---

Sequencing summary

1. Set up (avatar, header, bio, handle). Pin the longer bio alt as the first tweet.
2. Follow Section 4, start replying and resharing Section 6's first list.
3. Post value (Section 5) over several weeks, no selling, no study.
4. Warm the builder audience with the Claude Code thread (Section 7) once engagement is real.
5. Drop the study (Section 8) onto a warm audience, link to the live study page.
6. From there, repeat the value-first rhythm and let the next study installment compound on the following you've built.

Nothing here posts itself. All of it waits on Annette's SEO Believer X account.